Mouserat/qwen3-tts-0.6b-base-gguf

Qwen3-TTS-12Hz-0.6B-Base GGUF (llama.cpp official conversion)

Converted using the official conversion/qwen3tts.py in ggml-org/llama.cpp (mtmd Qwen3-TTS PR #26254), from Qwen/Qwen3-TTS-12Hz-0.6B-Base.

Files

• —qwen-talker-0.6b-base-Q8_0.gguf — backbone (talker), load with llama_model_load_from_file
• —qwen-tokenizer-12hz-f16.gguf — tokenizer/codec (mmproj), load with mtmd_init_from_file

Usage

llama-tts -m qwen-talker-0.6b-base-Q8_0.gguf -mm qwen-tokenizer-12hz-f16.gguf -p "Hello world" --output out.wav

Other GGUF conversions of this model floating around (e.g. koboldcpp-oriented exports) use non-standard metadata key names (qwen3-tts.* with hyphen, talker. subsections, general.file_type stored as string) that are incompatible with the official llama.cpp qwen3tts architecture loader. This repo uses the stock conversion script, so the metadata matches what llama.cpp expects out of the box.